Green AI: from nuclear energy to green prompting
- Philipp Hiestand
- Apr 12
- 1 min read
The use of GenAI and AI won't decrease in the upcoming years. Top companies like Google, Amazon and Microsoft are struggeling to implement their net zero emissions and AI Initiatives side by side.
In the short term those increased energy consumption are mostly covered with fossil energy production. Virginia data center valley which was the site of 70% internet traffic in 2019 was often powered by "dirty" energy. Dirty could be summarized with high CO2 emissions and high inefficiencies from primary energy to useful energy.
The long-term energy goals of those top companies are now also considering using their own nuclear fusion plants. But reestablishing plants could take some time and many administrative actions before those new datacenters will be active on the grid.
In the coming years those companies are looking into following practical application which are lie in eliminating energy inefficiences in transportation, transformation, computation, creation or requests. The list could be something like:
Selection more efficient energy sources like solar, wind or nuclear
Deploying specialized hardware which is more efficient than General Purpose CPU's
Using the fewer parameters model or best-fit models to execute simpler tasks
Train more efficiently aka #deekseek approach
Federated learning where dataset stay decentralized and only parameters of the deep neural network are reported back
Monitoring your footprint with tools like code carbon
(no finalized list)
As user we could use green prompting. This requires much more efforts while writing the prompt. Prompts need to be precise. Less ambiguity more efficient results.
Sources:
Comments